VectorStore

What it is

• A minimal in-memory vector store backed by Qdrant (via qdrant-client).
• Stores texts + metadata alongside provided embedding vectors and supports cosine-similarity search.

Public API

• class VectorStore(dimension: int = 1536)
  • Creates an in-memory Qdrant collection named "documents" with cosine distance and given vector dimension.
• VectorStore.add_texts(texts, metadatas=None, embeddings=None) -> List[str]
  • Upserts a batch of points containing:
    • payload["text"] = original text
    • additional payload fields from corresponding metadata dict
  • Requires embeddings (no embedding generation is performed).
  • Returns string IDs for inserted items.
• VectorStore.similarity_search(query_embedding, k: int = 4) -> List[dict]
  • Queries the collection with a provided embedding vector.
  • Returns a list of dicts: {"text": ..., "metadata": ..., "score": ...}.

Configuration/Dependencies

• Dependency: qdrant-client
  • Install: pip install qdrant-client
• Storage: in-memory Qdrant instance (QdrantClient(":memory:"))
• Collection:
  • name: "documents"
  • distance: cosine
  • vector size: dimension passed to constructor

Usage

from naas_abi_core.services.agent.beta.VectorStore import VectorStore

store = VectorStore(dimension=3)

texts = ["hello world", "goodbye world"]
metadatas = [{"source": "a"}, {"source": "b"}]
embeddings = [
    [1.0, 0.0, 0.0],
    [0.9, 0.1, 0.0],
]

ids = store.add_texts(texts=texts, metadatas=metadatas, embeddings=embeddings)
results = store.similarity_search(query_embedding=[1.0, 0.0, 0.0], k=2)

print(ids)
print(results)

Caveats

• add_texts(...) raises ValueError if embeddings is not provided.
• The Qdrant collection is always created in __init__ with a fixed name ("documents"); data is not persisted (in-memory only).
• IDs are auto-assigned sequential integers starting from 0 and returned as strings.